Data Visualization in R
Allison Shultz
The goal of this workshop is to teach the grammar of graphics in R, with a focus on ggplot2. The consistent grammar implemented in ggplot2 is advantageous both because it is easily extendible - that is you can both produce simple plots, but then develop them into complex publication-ready figures. In addition to the basic ggplot2 R package, many extensions for different types of data have been written using the same standardized grammar.
ggplot2 is part of the tidyverse package, and to make it easier to load our dataset and manipulate it prior to plotting, we will load the entire tidyverse package.
## Loading tidyverse: ggplot2
## Loading tidyverse: tibble
## Loading tidyverse: tidyr
## Loading tidyverse: readr
## Loading tidyverse: purrr
## Loading tidyverse: dplyr## Warning: package 'dplyr' was built under R version 3.4.2## Conflicts with tidy packages ----------------------------------------------## filter(): dplyr, stats
## lag(): dplyr, statsTo demonstrate how the different plots work, let’s import in a dataset into our environment. This is a dataset containing plumage color information for a family of birds called the tanagers.
setwd("~/Dropbox/Informatics/Workshops/VisualizationR/")
tanager_color <- read_csv("tanagercolordata.csv")## Parsed with column specification:
## cols(
## species = col_character(),
## sex = col_character(),
## subfamily = col_character(),
## habitat = col_character(),
## foraging_stratum = col_character(),
## avg_span = col_double(),
## volume = col_double(),
## avg_brill = col_double(),
## avg_chroma = col_double(),
## avg_sex_dich = col_double()
## )tanager_color## # A tibble: 688 x 10
## species sex subfamily habitat foraging_stratum
## <chr> <chr> <chr> <chr> <chr>
## 1 Acanthidops_bairdii Female Diglossinae F U
## 2 Acanthidops_bairdii Male Diglossinae F U
## 3 Anisognathus_igniventris Female Thraupinae F U/C
## 4 Anisognathus_igniventris Male Thraupinae F U/C
## 5 Anisognathus_lacrymosus Female Thraupinae F C
## 6 Anisognathus_lacrymosus Male Thraupinae F C
## 7 Anisognathus_melanogenys Female Thraupinae F U/C
## 8 Anisognathus_melanogenys Male Thraupinae F U/C
## 9 Anisognathus_notabilis Female Thraupinae F C
## 10 Anisognathus_notabilis Male Thraupinae F C
## # ... with 678 more rows, and 5 more variables: avg_span <dbl>,
## # volume <dbl>, avg_brill <dbl>, avg_chroma <dbl>, avg_sex_dich <dbl>You notice that our dataset is currently tidy, and in long-format. Each row represents an observation, in this case, information on plumage coloration for each sex of each species (avg_span = average contrast among color patches, volume = the volume of the avian tetrahedral space, larger values are more colorful, avg_brill = average brilliance across color patches, avg_chroma = average chroma, or saturation of each color patch, and avg_sex_dich = measures of sexual dichromatism, or difference in coloration between males and females for that species). As ggplot2 is part of the tidyverse, it works best with long-format data. In addition to these color measurements, we also have information on which subfamily each species belongs to, habitat preferences (F = closed habitat, or forest, N = open habitat), and where birds forage (U = understory, C = canopy, U/C = both). If you would like to know more, this dataset comes from this 2017 Evolution paper: http://onlinelibrary.wiley.com/doi/10.1111/evo.13196/full.
Basic grammar of ggplot
There are three key components that make up every ggplot: 1. data 2. aesthetic mappings (which variables in your data map to which visual properties) 3. geometric object (geom) function (a layer describing how to render each observation)
There are other optional components that control the visualization of the plots, but for now, we will focus on getting these three key elements down. The basic formula for these options is:
ggplot(data=<dataset>, aes(<mappings>)) + <geom_function>()
Let’s make a basic plot using this grammar. You can see that this is really not much more complicated than the base R plot function.
ggplot(data=tanager_color, aes(x=avg_brill, y=avg_chroma)) +
geom_point()
Notice that like when we use piping (%>%) with dplyr and can make our code easier to read by putting each element on a new line, we can do the same with ggplot after the +. That + is specifying that we want to add a layer to the plot. As you will see going forward, we can add more than one layer to add other data, statistical summaries, or metadata to make more complex plots.
Also, here we specified what x and y are, but they are used so commonly, that ggplot2 will always assume that the first aes() argument supplied is x, and the second is y, so we do not need to specify them moving forward.
Geoms
Geoms are the building blocks of ggplot2, and specify what type of plot you will be drawing. For example, you saw above that geom_point draws a scatterplot, or associates the x and y values with points.
We can divide geoms up into several different categories, such as individual geoms, which map each observation you provide to an element of the plot. Alternatively, collective geoms will group your points to summarize aspects of your data (e.g. a boxplot). Additionally, we can categorize geoms into how many dimensions they are mapping from 1D to 2D, or even 3D.
Another nice feature of ggplot2 is that we can save the data and aesthetic mappings to an object, and then call this object with different geoms or other layers. This can be very useful if you want to explore how different geoms work to visualize your data, or change subtle aspects of the plot.
1-D geoms
First, we will explore some of the geoms available for single variables. To facilitate comparing different plot types, let’s create a base ggplot2 object called brill with an aesthetic mapping the avg_brill values.
brill <- ggplot(tanager_color,aes(avg_brill))A simple way to explore the distribution of this variable is with a histogram.
brill + geom_histogram()## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
We likely will want to change the default values for the number of breaks, or the binwidth.
brill + geom_histogram(bins=15)
brill + geom_histogram(binwidth=0.005)
Density Plots Similar to histograms, we can make density plots, which smooths out the distribution of data instead of binning the data. We can change how smooth the estimate is with adjust. Values smaller than 1 result in less smoothing, and values greater than 1 result in more smoothing.
brill + geom_density()
brill + geom_density(adjust=0.1)
brill + geom_density(adjust=5)
Exercises
Pick another continuous variable in the dataset and produce a histogram. Try producing different versions with different binwidths.
Try adding geom_freqpoly() to your histogram plot in a second layer. What does it look like this plot is going? Does it match your histogram? If not, why?
Now try adding a density plot to your plot. What happens if you call the layers in a different order? What does the appearance suggest about how ggplot scales multiple layers?
q-q-Plots We can also produce q-q plots with a continuous variable to see how well distributions match. You can specify which distribution to test against, but the default is normal. One difference with the geom_qq() function is that it requires a different aesthetic mapping than we saw before - instead of x we need to specify a sample mapping.
ggplot(tanager_color,aes(sample=avg_brill)) + geom_qq()
Bar plots (discrete variables) With a discrete variable, we might want to use a barplot to explore the distribution of our data.
g <- ggplot(tanager_color, aes(habitat))
g + geom_bar()
2-D geoms
As we saw above with the scatterplot, we can use x and y mappings to plot two variables against each other.
When those two variables are continuous, we have a few other plot options in addition to a scatterplot. Notice that we can also add log-scaling to our basic aesthetic if that is required to visualize the relationship between our variables.
#If we round brilliance to 1 digit, we have many overlapping values
bv <- ggplot(tanager_color, aes(avg_brill, log(volume)))
bv + geom_point()
Rug plot
bv + geom_rug()
Smoothed line
bv + geom_smooth()## `geom_smooth()` using method = 'loess'## Warning: Removed 5 rows containing non-finite values (stat_smooth).
Continuous bivariate distributions
Sometimes if you have a lot of data, you run a risk of overplotting if you plot all points at once. There are a few ways to solve this issue. One would be to make your points partially transparent, and we will learn how to do that later in the workshop. Another way would be to plot the distribution of points. We can do that with geom_bin2d (heatmap of 2d bin counts), geom_density2d (2d density plot) or geom_hex(heatmap of hexagon counts).
bv + geom_bin2d()## Warning: Removed 5 rows containing non-finite values (stat_bin2d).
bv + geom_density2d()## Warning: Removed 5 rows containing non-finite values (stat_density2d).
#install.packages(hexbin) if you don't already have it installed.
bv + geom_hex()## Warning: Removed 5 rows containing non-finite values (stat_binhex).
Boxplot If you have a discrete x variable and continuous y variable, it might make sense to use boxplots or violinplots to compare distributions of points.
sv <- ggplot(tanager_color,aes(sex,log(volume)))
sv + geom_boxplot()## Warning: Removed 5 rows containing non-finite values (stat_boxplot).
Violin Plot A violin plot is similar plot, but shows the entire distribution of points (think a histogram turned on its side and mirrored)
sv + geom_violin()## Warning: Removed 5 rows containing non-finite values (stat_ydensity).
Discrete variables Sometimes, if you have two discrete variables it can be useful to compare the counts in all possible combinations of classes. We can use geom_count for that.
ggplot(tanager_color,aes(habitat,foraging_stratum)) + geom_count()
Line Functions Sometimes we have data that are best visualized by a line, such as time series data. To visualize this, we will use the built in economics dataset, and use geom_area (shade area below line) and geom_line(draw a straight line connecting points).
l <- ggplot(economics, aes(date,pop))
l + geom_area()
l + geom_line()
If you want to draw a line without basing it off of data, you can do with geom_abline (specify intercept and slope), geom_hline (will be a horizontal line, and you specify the y intercept), or geom_vline (will be a vertical line, and you specify the x intercept).
As an example, let’s spread the male and female measurements for avg_span so that we can plot them against each other. If males and females are equal, we would expect them to fall on a y=x line. Do you obsesrve that trend?
tcol_span_spread <- tanager_color %>% select(species,subfamily,sex,foraging_stratum,habitat,avg_span) %>% spread(sex,avg_span)
ggplot(tcol_span_spread,aes(Male,Female)) +
geom_point() +
geom_abline(aes(intercept=0,slope=1))## Warning: Removed 4 rows containing missing values (geom_point).
Exericses
Produce a smoothed plot of avg_span by avg_brill. However, use a linear model to estimate the smoothing. Try both with and without standard error. Try adding a scatterplot layer.
Compare log(volume) among different subfamilies. What type of plotting do you think is most informative?
Plot male versus female avg_brill. Add a y=x line. How might you also visualize the distribution of data on the same plot using the geoms we have already explored?
Grouping
In some situations, it might make sense to compare data among different groups of points. Using group, it is possible to apply geoms that summarize (e.g. geom_smooth) to groups of points. This can give a general idea of how trends might vary.
ggplot(tcol_span_spread,aes(Male,Female,group=subfamily)) +
geom_point() +
geom_abline() +
geom_smooth(method="lm",se=FALSE)## Warning: Removed 4 rows containing non-finite values (stat_smooth).## Warning: Removed 4 rows containing missing values (geom_point).
In order to actually identify which groups of points belong to what, it is necessary to use col instead of group. Col will group by default.
ggplot(tcol_span_spread,aes(Male,Female,col=subfamily)) +
geom_point() +
geom_abline() +
geom_smooth(method="lm",se=FALSE)## Warning: Removed 4 rows containing non-finite values (stat_smooth).## Warning: Removed 4 rows containing missing values (geom_point).
Fill can be used to specify fill colors for objects like boxplots.
ggplot(tanager_color,aes(foraging_stratum,avg_brill,fill=habitat)) +
geom_boxplot()
However, if the fill object is not hierarchical, it will actually produce side-by-side plots of the fill categories as well
ggplot(tanager_color,aes(foraging_stratum,avg_brill,fill=sex)) +
geom_boxplot()
Note that you do not have to put the group, col, or fill variable in the original aes map, but can put it for individual graph components to build summary statistics.
ggplot(tcol_span_spread,aes(Male,Female)) +
geom_point(aes(col=foraging_stratum)) +
geom_smooth(aes(col=habitat),method="lm",se = FALSE)## Warning: Removed 4 rows containing non-finite values (stat_smooth).## Warning: Removed 4 rows containing missing values (geom_point).
Statistical transformations (Stats)
You can summarize data using statistical transformations, or stats. Without realizing it, we actually have already bee using stat functions to build our plots. This is because the summary geoms (e.g. geom_boxplot), uses these functions natively (e.g. stat_boxplot).
The stat functions behind the various geoms that summarize data are useful to know about, as they will provide more detail and information in the documentation. However, there are some stats that are separate from the basic geom functions we have been learning so far. For example
stat_function= computes y values from a function of x values.stat_unique= removes duplicate rowsstat_summary= summarises y values at distinct x values
There are two ways to use the stat functions.
- Add a
stat_()function and override the default geom.
subfam_span <- ggplot(tanager_color,aes(subfamily,avg_span))
subfam_span + geom_point() +
stat_summary(geom="point", fun.y= "mean", color = "red", size=4)
- Add a
geom_()function and override the default stat.
subfam_span + geom_point() +
geom_point(stat="summary", color = "red", size=4)## No summary function supplied, defaulting to `mean_se()
One time you likely will want to override the default stat options is when you want to plot summary variables you have calculated outside of ggplot2 (e.g. as a barplot). In that case, you would specify stat = "identity", as the default stat for a barplot is stat = "count".
tcol_span_mean <- tanager_color %>%
group_by(subfamily) %>%
summarize(mean_span = mean(avg_span), sd_span = sd(avg_span))
ggplot(tcol_span_mean,aes(subfamily,mean_span)) +
geom_bar(stat="identity")
Error
ggplot2 has several ways to display error, whether you calculated it yourself (and for example have se values), or whether or not you want to use some of the basic stats available within ggplot2.
The default function for stat_summary is actually mean_se, and will return both mean and standard error calculations.
subfam_span + stat_summary()## No summary function supplied, defaulting to `mean_se()
There are quite a few options for plotting errorbars in addition to the simple point with a range we saw above.
Crossbars
subfam_span + stat_summary(geom="crossbar")## No summary function supplied, defaulting to `mean_se()
Errorbars
subfam_span + stat_summary(geom="errorbar")## No summary function supplied, defaulting to `mean_se()
Linerange
subfam_span + stat_summary(geom="linerange")## No summary function supplied, defaulting to `mean_se()
Finally, you can actually give it any functions you would like, by specifying fun.y, fun.ymin, and fun.ymax. For example, if you want to show mean +- sd:
subfam_span +
stat_summary(fun.y = mean,
fun.ymin = function(x) mean(x) - sd(x),
fun.ymax = function(x) mean(x) + sd(x),
geom = "pointrange")
Exercies:
Plot average brilliance (avg_brill) by foraging stratum. Set the initial ggplot function of data and aesthetic mapping to an object called brill_fs. Make sure to include col=foraging_stratum in the inital aesthetic map. Plot by adding a boxplot layer, and fill each boxplot by overall habitat.
Now, use your brilL_fs plot, but instead show the mean and standard error for each category. Show the mean and standard error via crossbar.
Finally, calculate the mean and standard deviation for average brilliance in each foraging stratum outside of ggplot. Plot errorbars of the standard deviation as you calculated them in addition to a barplot of the mean. Make sure the bars are filled by foraging stratum.
Answers:
#1
brill_fs <- ggplot(tanager_color, aes(foraging_stratum,avg_brill,col=foraging_stratum))
brill_fs + geom_boxplot(aes(fill=habitat))
#2
brill_fs + stat_summary(geom = "crossbar")## No summary function supplied, defaulting to `mean_se()
#3
tcol_brill_meansd <- tanager_color %>% group_by(foraging_stratum) %>%
summarize(mean = mean(avg_brill), sd = sd(avg_brill))
ggplot(tcol_brill_meansd,aes(foraging_stratum,mean,fill=foraging_stratum)) +
geom_bar(stat="identity") +
geom_errorbar(aes(foraging_stratum,ymin=mean-sd,ymax=mean+sd))
Plot Appearances
Scales
Scales are essential parts of how ggplot2 functions, and are always present whether or not you are aware of them. Scales take your data and map it to aesthetics. They give control over how the mapping works (e.g. size, color, position), axes and legends.
Scales are named according to: 1. scale_ 2. axis (e.g. x_ or color_) 3. variable type (e.g. continuous or discrete)
So, when we plot: ggplot(tanager_color,aes(foraging_stratum,avg_brill,col=habitat)) + geom_point()
We are actually plotting: ggplot(tanager_color,aes(foraging_stratum,avg_brill,col=habitat)) + geom_point() + scale_x_discrete() + scale_y_continuous() + scale_color_discrete()
You can actually change attributes of the scale by specifying the name of the scale. You can even add mathmatical equations by using the quote() function. Other attributes like data transformations can be added as well.
brill_fs_hab <- ggplot(tanager_color,aes(foraging_stratum,avg_brill,col=habitat))
brill_fs_hab + geom_point() +
scale_x_discrete(name="foraging stratum") +
scale_y_continuous(name="average brilliance",trans="log10") +
scale_color_discrete(name="Hab")
brill_fs_hab + geom_point() +
scale_x_discrete(name="foraging stratum") +
scale_y_continuous(name=quote(math-something^2),trans="log10")
As this is the most common attribute to change, ggplot2 helpfully also has implemented xlab(), ylab() and labs() functions as well to simplify changing axis names.
brill_fs_hab + geom_point() +
xlab("foraging stratum") +
ylab("average brilliance") +
labs(title="Great title!", subtitle="Awesome subtitle!")
Another visual attirbute you can update are the names and values of the tick marks. The breaks attribute specifies where the tick marks should appear, and the labels attribute specifies what they should be called. You can also rename values in a discrete scale with the labels attribute. The minor breaks can be set by the minor_breaks attribute.
brill_fs_hab + geom_point() +
scale_y_continuous(breaks=c(0.05,0.1,0.15,0.2,0.25,0.30,0.35),minor_breaks=NULL) +
scale_x_discrete(labels= c("-"="missing", "C"="canopy", "N"="open", "U"="understory", "U/C"="understory/canopy"))
#Shortcut
#brill_fs_hab + geom_point() +
# scale_y_continuous(breaks=seq(0.05,0.35,by = 0.05))Finally, you might sometimes want to change the limits of the x or y axis. Within the scale_ fuction this can be done by specifying limits = c(lowerlimit,upperlimit). Like with labels, this can also be done with special xlim(lowerlimit,upperlimit) or ylim functions.
brill_fs_hab + geom_point() +
scale_y_continuous(breaks=c(0.05,0.1,0.15,0.2,0.25,0.30,0.35),minor_breaks=NULL, limits=c(0,0.25)) +
scale_x_discrete(labels= c("-"="missing", "C"="canopy", "N"="open", "U"="understory", "U/C"="understory/canopy"))## Warning: Removed 28 rows containing missing values (geom_point).
Legends
The term for legends in ggplot2 is guide. Another nice improvement of ggplot2 over base R graphics is that guides are automatically generated for you based on aesthetic mappings. If you have ever tried to create legends in base R, you know that you have to specify all aspects by hand, and this can much more easily lead to errors.
You will notice that the guide will draw symbols from multiple layers if necessary, and the symbol will correspond to the layer type and appearance. Points will appear as points, lines as lines, and fill color as a colored rectangle.
brill_fs_hab + geom_point(size=4,pch=8) +
geom_point(aes(col=foraging_stratum), size=1) +
scale_y_continuous(breaks=c(0.05,0.1,0.15,0.2,0.25,0.30,0.35),minor_breaks=NULL) +
scale_x_discrete(labels= c("-"="missing", "C"="canopy", "N"="open", "U"="understory", "U/C"="understory/canopy"))
If you don’t want a layer to appear in a legend, you can add show.legend = FALSE and it will not be added.
brill_fs_hab + geom_point(show.legend=FALSE)
It is possible to change the position and orientation of the guide within the theme setting (more on this later). Within the theme() call, you simply set legend.position equal to right, left, bottom, or none. You can also control if the keys are laid out horizontally or vertically with legend.direction equals vertical or horizontal.
brill_fs_hab + geom_point() +
scale_color_discrete(labels=c("-"="missing","F"="forest","N"="open"))+
theme(legend.position = "bottom", legend.direction = "vertical")
Sometimes you don’t want the legend to exactly reflect what is currently in a plot. For example, if you use transparency in your color (we will learn more about this below!), you might want the legend to reflect the color without transparency. You can override this parameter in the guides() function, which is kind of like labs(). Alternatively, in the scale call, you can change inputs in with the guide=guide_legend() (for continuous or discrete characters) or guide=guide_colorbar() (continuous color gradients)
brill_fs_hab + geom_point(alpha = 0.2)
brill_fs_hab + geom_point(alpha = 0.2) +
guides(color=guide_legend(override.aes = list(alpha=1)))
Exercises:
Change the labels in the legend for the brill_fs_hab plot to missing, forest and open. Try changing the legend.position to a numeric vector of length 2 (c(0,1)), where the two numbers vary from 0 to 1. Use these numbers to position the legend in a place that is not overlapping any points.
On the same plot, Look up the help for guide_legend() and change the title of the legend, reverse the order of the legend elements, and make it so the legend squares are on more than 1 row.
Create a new scatterplot of average color span versus the log of color volume, colored by sex. Change the axis labels to represent the actual trait names, and change the “Female” and “Male” labels in the legend to be lowercase. How does the legend change when you add a linear model line to the points?
Answers:
#1
brill_fs_hab + geom_point() +
scale_color_discrete(labels=c("-"="missing","F"="forest","N"="open"))+
theme(legend.position = c(0.75,0.8))
#2
brill_fs_hab + geom_point() +
scale_color_discrete(labels=c("-"="missing","F"="forest","N"="open"),guide=guide_legend(title="test",reverse=TRUE,nrow=2))+
theme(legend.position = c(0.75,0.8))
#3
ggplot(tanager_color,aes(log(avg_span),log(volume),col=sex)) + geom_point() + geom_smooth(method="lm")## Warning: Removed 5 rows containing non-finite values (stat_smooth).
Colors and shapes
Continuous
Generally visualizing continuous colors can be performed with a color gradient.
It is possible to change these gradient values with the scale_ function.
span_vol <- ggplot(tanager_color,aes(log(avg_span),log(volume)))
span_vol + geom_bin2d()## Warning: Removed 5 rows containing non-finite values (stat_bin2d).
span_vol + geom_bin2d() +
scale_fill_gradient(low="white",high="black")## Warning: Removed 5 rows containing non-finite values (stat_bin2d).
It is possible to set a 3 color gradient (e.g. for differences in gene expression, negative, neutral, positive) with ‘scale_fill_gradient2’. With our dataset, we will artificially set a midpoint just for visualization purposes. You also can use scale_fill_gradientn to create a custom gradient with as many different colors as you want. Here we will use a few color vectors that already exist in R, but you can input any custom set of colors.
span_vol + geom_bin2d() +
scale_fill_gradientn(colors = terrain.colors(10))## Warning: Removed 5 rows containing non-finite values (stat_bin2d).
span_vol + geom_bin2d() +
scale_fill_gradientn(colors = rainbow(10))## Warning: Removed 5 rows containing non-finite values (stat_bin2d).
Keep in mind that you can also choose the color for missing data in these gradients with the na.value argument. By default this is set to grey.
Discrete
Natively in ggplot2 there are four color scales, that are supposed to be evenly spaced in the color wheel. You can change the chroma (c), range of hues (h), or lightness (l)
g <- ggplot(tanager_color, aes(foraging_stratum,fill=foraging_stratum))
g + geom_bar()
g + geom_bar() + scale_fill_hue(c=40)
g + geom_bar() + scale_fill_hue(h=c(180,300))
g + geom_bar() + scale_fill_hue(l=40)
g + geom_bar() + scale_fill_hue(l=60)
g + geom_bar() + scale_fill_hue(l=80)
g + geom_bar() + scale_fill_hue(l=100)
However, ggplot2 has built in all of the color palettes form “ColorBrewer” (http://colorbrewer2.org), which can be a great option to choose. You can choose this with scale_color_brewer or scale_fill_brewer, and set the name of the color palette with the palette argument. The direction argument can be used to change the order of the colors.
g + geom_bar() + scale_fill_brewer(palette="Set1")
g + geom_bar() + scale_fill_brewer(palette="Set1",direction=-1)
If you are looking for a greyscale, you can use scale_color_grey(), which maps discrete data from light to dark. To change the direction, or where you start on the scale, you can specify start and end.
g + geom_bar() + scale_fill_grey()
g + geom_bar() + scale_fill_grey(start=0,end=0.75)
g + geom_bar() + scale_fill_grey(start=1,end=0)
If you have your own color palettes, you can use scale_color_manual(). This can be very useful if you want to set specific colors for specific classes of data (e.g. to be consistent among figures in a paper).
g + geom_bar() + scale_fill_manual(
values=c("-"="grey","C"="red","U"="yellow","U/C"="orange","N"="blue")
)
Finally, for all types of colors, you can use the alpha option to change the transparency of a color. This can be advantageous when plotting dense distributions of points, or overlaying objects (e.g. histograms).
ggplot(tanager_color,aes(avg_span,fill=sex, color=sex)) +
geom_density(alpha=0.2)
Exercises:
Create a boxplot of average chroma by subfamily using a color palette available in the color brewer. Remember to change both the outlines and the fill color.
Change the transparency of the fill color, but make sure the legend shows the full color without transparency.
Positioning
By changing the basic positioning of aspects of the data, it is possible to create very different plots. Here, we will use barplots to demonstrate:
ggplot(tanager_color,aes(habitat,fill=sex)) +
geom_bar()
ggplot(tanager_color,aes(habitat,fill=sex)) +
geom_bar(position="dodge")
ggplot(tanager_color,aes(habitat,fill=sex)) +
geom_bar(position="fill")
You can also use position elements on other types of geoms, such as points. When you have many overlapping points for example, it can be useful to use position=jitter to add random noise to each x and y value.
brill_fs_hab + geom_point()
brill_fs_hab + geom_point(position="jitter")
Coordinate System
The default coordinate system used by ggplot2 is a cartesian coordinates, where x and y combine to produce a position on the plot.
It is possible to change the coordinates, either in a linear or non-linear fashion.
In a linear fashion, you can use the coord_cartesian function to change the limits of your plot (essentially zooming in). This can be advantageous because the data outside the limits are not just thrown away as we saw with setting axis limits in the scale functions.
brill_fs_hab + geom_point(alpha=0.2) +
stat_summary(col="black")## No summary function supplied, defaulting to `mean_se()
brill_fs_hab + geom_point(alpha=0.2) +
scale_y_continuous(limits=c(0,0.15)) +
stat_summary(col="black")## Warning: Removed 408 rows containing non-finite values (stat_summary).## No summary function supplied, defaulting to `mean_se()## Warning: Removed 408 rows containing missing values (geom_point).
brill_fs_hab + geom_point(alpha=0.2) +
coord_cartesian(ylim=c(0,0.15)) +
stat_summary(col="black")## No summary function supplied, defaulting to `mean_se()
Another coordinate function is coord_flip(), which flips the y and x axes. This can be useful for visualization purposes, or to change the assumptions of statistical models used in summary functions.
brill_fs_hab + geom_point(alpha=0.2) +
stat_summary(col="black") +
coord_flip()## No summary function supplied, defaulting to `mean_se()
coord_fixed can make sure that the ratio of x to y axes are equal, no matter the shape of the output. This is useful if you want a relationship to remain a certain way, no matter how the plot comes out.
It is also possible to translate linear coordinates into nonlinear ones. For example, it is possible to transform linear coordinates into polar coordinates You can even do this to create a pie chart (I am not necessarily recommending pie charts.)
brill_fs_hab + geom_point(alpha=0.2) +
coord_polar("y")
brill_fs_hab + geom_point(alpha=0.2) +
coord_polar("x")
ggplot(tanager_color,aes(habitat,fill=sex)) +
geom_bar() +
coord_polar()
ggplot(tanager_color,aes(factor(1),fill=habitat)) +
geom_bar(position="fill") +
coord_polar(theta="y")
Finally, coord_map also can quickly turn long and lat data into coordinates for mapping.
#Cartesian
ggplot(map_data("world"),aes(long,lat,group=group)) +
geom_polygon(fill="white", color="black")##
## Attaching package: 'maps'## The following object is masked from 'package:purrr':
##
## map
#Aspect ratio approx.
ggplot(map_data("world"),aes(long,lat,group=group)) +
geom_polygon(fill="white", color="black") +
coord_quickmap()
#Other aspect ratios
ggplot(map_data("world"),aes(long,lat,group=group)) +
geom_polygon(fill="white", color="black") +
coord_map("ortho")
ggplot(map_data("world"),aes(long,lat,group=group)) +
geom_polygon(fill="white", color="black") +
coord_map("stereographic")
Faceting
Facetting can be the most useful to generate plots of the same variables, but for different subsets of the data.
There are two options, facet_wrap and facet_grid. facet_wrap will wrap all of the variables into a single horizontal line, broken into however many rows you specify.
ggplot(tanager_color,aes(log(volume))) +
geom_histogram(aes(fill=subfamily)) +
facet_wrap(~subfamily)## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.## Warning: Removed 5 rows containing non-finite values (stat_bin).
facet_grid is nice because it can be used to show either one or two variables. You specify via: - .~var1 will spread the values of var1 across columns. - var1~. will spread the values of var1 across rows. - var1~var2 will spread var1 across columns and var2 across rows.
ggplot(tanager_color,aes(log(volume))) +
geom_histogram(aes(fill=sex)) +
facet_grid(.~sex)## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.## Warning: Removed 5 rows containing non-finite values (stat_bin).
ggplot(tanager_color,aes(log(volume))) +
geom_histogram(aes(fill=sex)) +
facet_grid(sex~.)## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.## Warning: Removed 5 rows containing non-finite values (stat_bin).
ggplot(tanager_color,aes(log(volume))) +
geom_histogram(aes(fill=sex)) +
facet_grid(sex~habitat)## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.## Warning: Removed 5 rows containing non-finite values (stat_bin).
You can also combine facetting and groups to learn different things about your data.
tcol_span_spread2 <- tcol_span_spread %>% select(-foraging_stratum)
ggplot(tcol_span_spread,aes(Female,Male)) +
geom_point(data=tcol_span_spread2,color="grey") +
geom_point(aes(color=foraging_stratum)) +
facet_wrap(~foraging_stratum)## Warning: Removed 20 rows containing missing values (geom_point).## Warning: Removed 4 rows containing missing values (geom_point).
Themes
All non-data related elements of the plot can be controlled by the theme system.
Theme elements specify which non-date element you want to change (e.g. plot.title, axis.ticks.x).
Theme element functions describes the visual properties of an element. For example, element_text changes textual elements like the plot.title.
There are a number of different built in themes, other than the default, which is theme_grey(). Some are:
brill_fs_hab + geom_boxplot(aes(fill=habitat),alpha=0.2) +
theme_grey()
brill_fs_hab + geom_boxplot(aes(fill=habitat),alpha=0.2) +
theme_bw()
brill_fs_hab + geom_boxplot(aes(fill=habitat),alpha=0.2) +
theme_linedraw()
brill_fs_hab + geom_boxplot(aes(fill=habitat),alpha=0.2) +
theme_dark()
brill_fs_hab + geom_boxplot(aes(fill=habitat),alpha=0.2) +
theme_light()
brill_fs_hab + geom_boxplot(aes(fill=habitat),alpha=0.2) +
theme_minimal()
brill_fs_hab + geom_boxplot(aes(fill=habitat),alpha=0.2) +
theme_classic()
brill_fs_hab + geom_boxplot(aes(fill=habitat),alpha=0.2) +
theme_void()
Here is an example of modifying some elements of a theme:
brill_fs_hab + geom_boxplot(aes(fill=habitat),alpha=0.2) +
theme_bw() +
theme(
panel.grid.major.x=element_blank(),
axis.text = element_text(size=14)
)
Saving and exporting plots
To save your plots, there are two options:
- Use the standard R approach, where you open a graphics device, create plots, and close it.
pdf(file="Testplot.pdf")
brill_fs_hab + geom_boxplot(aes(fill=habitat),alpha=0.2) +
theme_bw() +
theme(
panel.grid.major.x=element_blank(),
axis.text = element_text(size=14),
axis.title = element_text(size=14)
)
dev.off()## quartz_off_screen
## 2- Use the special
ggsavefucntion to save a plot after you have created it interactively.
brill_fs_hab + geom_boxplot(aes(fill=habitat),alpha=0.2) +
theme_bw() +
theme(
panel.grid.major.x=element_blank(),
axis.text = element_text(size=14),
axis.title = element_text(size=14)
)
ggsave("Testplot2.pdf")## Saving 7 x 5 in imageA few options in ggsave that can be hand are to specify the dimensions with the width and height arguments (default in inches), and that the path extension you use will automatically select the correct graphics device (e.g. .eps, .pdf, .svg, .png, .jpg, .tiff, .bmp).
Exercises:
With the above plot, change the x-axis labels to the full names “canopy”, “understory”, etc. Change the orientation of these labels to diagonal (look up options with ?theme.)
Create a facet plot of density plots of average chroma by foraging stratum and sex. Make sure to color by sex. Change the x axis label to be “average chroma”
With the plot above, change the theme to dark, and increase the axis font label size to 16 and make them bold.
Save as a jpg
Answers
#1
brill_fs_hab + geom_boxplot(aes(fill=habitat),alpha=0.2) +
scale_x_discrete(labels=c("-"="missing", "C"="canopy", "N"="open", "U"="understory", "U/C"="understory/canopy")) +
theme_bw() +
theme(
panel.grid.major.x=element_blank(),
axis.text = element_text(size=14),
axis.title = element_text(size=14),
axis.text.x = element_text(angle=45,vjust=0.5)
)
ggplot(tanager_color,aes(avg_chroma)) +
geom_density(aes(fill=sex)) +
facet_grid(foraging_stratum~sex) +
xlab("average chroma") +
theme_dark() +
theme(text = element_text(size=14),
axis.title = element_text(size=16,face="bold"))
ggplot extensions
Many extensions are available for ggplot, and now that you understand the grammar of graphics, you will be well-equipped to implement them. You can find the officially supported extensions here: http://www.ggplot2-exts.org, and there are many more, like ggtree or gggene.